File size: 81,917 Bytes
6fa4bc9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 |
{
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T01:55:18.694043Z"
},
"title": "Talk to Papers: Bringing Neural Question Answering to Academic Search",
"authors": [
{
"first": "Tianchang",
"middle": [],
"last": "Zhao",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "SOCO AI",
"location": {}
},
"email": "tonyzhao@soco.ai"
},
{
"first": "Kyusong",
"middle": [],
"last": "Lee",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "SOCO AI",
"location": {}
},
"email": "kyusongl@soco.ai"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "We introduce Talk to Papers 1 , which exploits the recent open-domain question answering (QA) techniques to improve the current experience of academic search. It's designed to enable researchers to use natural language queries to find precise answers and extract insights from a massive amount of academic papers. We present a large improvement over classic search engine baseline on several standard QA datasets, and provide the community a collaborative data collection tool to curate the first natural language processing research QA dataset via a community effort.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "We introduce Talk to Papers 1 , which exploits the recent open-domain question answering (QA) techniques to improve the current experience of academic search. It's designed to enable researchers to use natural language queries to find precise answers and extract insights from a massive amount of academic papers. We present a large improvement over classic search engine baseline on several standard QA datasets, and provide the community a collaborative data collection tool to curate the first natural language processing research QA dataset via a community effort.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Natural language processing (NLP) is one of the fastest growing field in computational linguistics and artificial intelligence, e.g. ACL has experienced a 140% growth from 2017 (1419 submissions) to 2020 (3429 submissions). Plus, there are more than 4000 pre-prints published at ArXiv in 2019. As a result, it has become increasingly stressful for researchers to keep up with the evolution of new methods. Today, the common way for researchers to find relevant papers is via searching keywords in Google Scholar 2 or Semantic Scholar 3 . Although these search engines are great at curating all the papers, they are limited in the following ways: (1) they are based on classic information retrieval methods, and do not handle natural language queries well, e.g. what effects can we get from label smoothing? (2) they are designed to find relevant documents (title and abstract) instead of direct answers to users' questions. Often researchers are looking for answers on specific research questions, e.g, how to prevent posterior collapse in VAE? or how much is it to label sentences via crowdsourcing? With current search engine, it requires one to read several papers to find these answer. Therefore, it is necessary to create better tools for researchers to find answers from the scientific publications in a more efficient manner.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Meanwhile, machine reading comprehension (MRC), aka question answering (QA) has advanced significantly. Pretrained and then fine-tuned transformer models (Devlin et al., 2018) have surpassed human performance on a number of datasets, e.g. SQuAD (Rajpurkar et al., 2016) . Further, Chen et al. (2017) extended single document MRC to machine reading at scale (MRS), combining the challenges of document retrieval with reading comprehension. Their open-domain QA system is able to find precise answers from millions of unstructured documents using natural language queries and has successfully been applied to the entire Wikipedia which contains more than 5 million articles.",
"cite_spans": [
{
"start": 154,
"end": 175,
"text": "(Devlin et al., 2018)",
"ref_id": "BIBREF3"
},
{
"start": 245,
"end": 269,
"text": "(Rajpurkar et al., 2016)",
"ref_id": "BIBREF14"
},
{
"start": 281,
"end": 299,
"text": "Chen et al. (2017)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The goal of Talk to Papers is to create a new way of finding answers from scientific publications and advance QA research. Concretely, we first adapted MRS techniques to create a conversational search portal that enable users to ask natural language questions to find precise answers and extract insights from the last 3 year papers published in top-tier NLP conferences, including ACL, NAACL, EMNLP and etc. Second, an initial corpus on these papers is collected and will be released as a publicly available dataset for QA research. We also developed a collaborative annotation toolkit that enable any researcher to contribute to this dataset so that more potential answers from these papers can be annotated. The annotation results will be fed back to the QA corpus after manual validation. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Developing question answering system with text knowledge base has been studied for decades (Voorhees and Tice, 1999) . Many of the classic system as well as recent MRC-based opendomain QA systems have relied a pipeline approach (Ferrucci et al., 2010; Chen et al., 2017; : (1) a information retrieval-based retriever module first finds relevant passages from all the documents and then (2) a reader component (computationally more expensive) extracts precise answer spans from the retrieved passages. has shown that using paragraphs as the unit of passage outperform sentences or documents. Lee et al. (2019a) proposes a trainable first-stage retriever that improves the recall performance. Pipeline-based system often suffer from error propagation (Zhao and Eskenazi, 2016) . Thus another line of research has been finding an end-toend approach that enable precise-answer extraction from the entire dataset instead of only the output from the first-stage retriever. Seo et al. (2019) introduced the phrase level representation model that index every potential answer span as vector representation and exploited approximate nearest neighbour (ANN) methods to retrieve the final answer span directly from a large vector index (Slaney and Casey, 2008) . Ahmad et al. (2019) argued that phrase-level answer may not always be required or preferred. Instead they proposed to find the right \"sentence\" as an answer from large body of text, and used universal sentence encoder (Cer et al., 2018) to retrieved the correct sentence given a question.",
"cite_spans": [
{
"start": 91,
"end": 116,
"text": "(Voorhees and Tice, 1999)",
"ref_id": "BIBREF19"
},
{
"start": 228,
"end": 251,
"text": "(Ferrucci et al., 2010;",
"ref_id": "BIBREF5"
},
{
"start": 252,
"end": 270,
"text": "Chen et al., 2017;",
"ref_id": "BIBREF2"
},
{
"start": 591,
"end": 609,
"text": "Lee et al. (2019a)",
"ref_id": "BIBREF10"
},
{
"start": 749,
"end": 774,
"text": "(Zhao and Eskenazi, 2016)",
"ref_id": "BIBREF21"
},
{
"start": 967,
"end": 984,
"text": "Seo et al. (2019)",
"ref_id": "BIBREF16"
},
{
"start": 1225,
"end": 1249,
"text": "(Slaney and Casey, 2008)",
"ref_id": "BIBREF17"
},
{
"start": 1252,
"end": 1271,
"text": "Ahmad et al. (2019)",
"ref_id": "BIBREF0"
},
{
"start": 1470,
"end": 1488,
"text": "(Cer et al., 2018)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Our approach follows the sentence-level QA system from (Ahmad et al., 2019) for two reasons: (1) answers to many research questions cannot be cov-ered in a short phrase-level span, and a sentence answer can provide more context to deliver relevant solutions. (2) our preliminary study found that it is important to have a trainable retriever that goes beyond TF-IDF keyword matching to ensure enough recall in the paper domain. Nonetheless, we keep a machine reader as optional post-process to extract phrase-level span from the sentences.",
"cite_spans": [
{
"start": 55,
"end": 75,
"text": "(Ahmad et al., 2019)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "3 The Proposed QA Toolkits: SOCO",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "We first introduce SOCO (Search Oriented COnversation), which we used to build our Talk to Paper. SOCO 4 is an answer-engine platform that enables developers to easily build universal question answering systems with unstructured documents as its knowledge base. Figure 1 shows the overall architecture of SOCO engine. It's designed to enable users to use natural language queries to find precise answers and extract insights from massive amount of text data. The typical workflow is as following:",
"cite_spans": [],
"ref_spans": [
{
"start": 262,
"end": 270,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "1. Split documents into sentences and convert each sentence with its context into semantic index (i.e. a collection of answer embedding, sparse features and other semantic features).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "2. Use natural language to query the index, which first converts the query into semantic embedding and then retrieves all the high probable answers.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "3. (Optional) Run machine reader to narrow down to phrase-level answers.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "We define a frame to be the basic building block of SOCO. Each frame contains N , where N is the total number of frames, a i is the potential answer sentence, c i is surrounding context of a i , and Q i is a set of questions that are manually/synthetically associated with the answer a i . Note that Q i is optional and often only a small set of frames are manually labelled. There are two neural network models involve in SOCO QA. The first model h a = F a (a, c) is an answer encoder that takes both the answer sentence and its surround context to create a contextsensitive answer embeddings h a . The second model is a question encoder h q = F q (q) that takes a query as input and maps it to a question embedding vector of the same size. Last, we define the relevance between a query and an answer frame to be s = cos(h a , h q ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "SOCO-Question Answering",
"sec_num": "3.1"
},
{
"text": "f i = [a i , c i , Q i ] i \u2208",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "SOCO-Question Answering",
"sec_num": "3.1"
},
{
"text": "These two models are trained jointly via supervised learning on existing QA dataset with cross entropy loss, i.e.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Training",
"sec_num": "3.1.1"
},
{
"text": "L = \u2212 j\u2208Jpos log(s j ) \u2212 j\u2208Jneg log(1 \u2212 s j ) (1)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Training",
"sec_num": "3.1.1"
},
{
"text": "where J pos is the set of ground truth questionanswer pairs, and J neg is the set of negative examples with randomly sampled noisy answers.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Training",
"sec_num": "3.1.1"
},
{
"text": "Given these two models and a set of frames, SOCO creates an index by encoding both the answers and annotated questions using F a and F q and save the resulting vectors D for nearest neighbour retrieval. Since F q and F a are trained to map the input text into the same embedding space, questionto-answer relevance and question-to-question relevance can be computed and compared in the same scale via cosine similarity.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Training",
"sec_num": "3.1.1"
},
{
"text": "At inference stage, SOCO first encodes the input query q via h q = F q (q ). Then each answer in the QA-index is scored by the cosine similarity between the query embedding and each answer embedding with a weighted auxiliary score from classic BM25 score (Robertson et al., 2009) .",
"cite_spans": [
{
"start": 255,
"end": 279,
"text": "(Robertson et al., 2009)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "3.1.2"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "y i = cos(h i , h q ) + \u03b1BM25(a i , q ) i \u2208 |D|",
"eq_num": "(2)"
}
],
"section": "Inference",
"sec_num": "3.1.2"
},
{
"text": "Note that an answer may have more than one vectors in the index because of the optional annotated question Q set in the frame, i.e. [h a , {h q }] q \u2208 Q. We merge the scores for the same answers via max pooling. Eventually, SOCO outputs the top K answers based on the final score.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "3.1.2"
},
{
"text": "One common issue for new users to use question answering system is that they may not know what kind of questions they can ask. Question generation (Du et al., 2017) is one of the solutions to this issue by suggesting users potential questions they may enter. Concretely, we created a question generator by fine-tuning a GPT-2 language model (Radford et al., 2019) . We train the model by concatenating question answers pairs [a, q] from QA corpus and fine tune a GPT-2 by maximizing the conditional log likelihood log P (q|a). The results questions are added to the Q set of each frame and is used to provide auto completion and FAQs in the search interface.",
"cite_spans": [
{
"start": 147,
"end": 164,
"text": "(Du et al., 2017)",
"ref_id": "BIBREF4"
},
{
"start": 341,
"end": 363,
"text": "(Radford et al., 2019)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "SOCO-Question Generation",
"sec_num": "3.2"
},
{
"text": "The SOCO python package (soco-core-python) is publicly available and can be installed as a Python package by running pip3 install soco-core-python. Internally, SOCO uses Elastic search (ES) (Gormley and Tong, 2015) as its index backbone. ES has built-in support for vector search, BM25 as well as context filtering. The answer and question encoder are trained on publicly available QA datasets, including SQuAD (Rajpurkar et al., 2016) , Natural Questions and MSMARCO (Nguyen et al., 2016) .",
"cite_spans": [
{
"start": 411,
"end": 435,
"text": "(Rajpurkar et al., 2016)",
"ref_id": "BIBREF14"
},
{
"start": 468,
"end": 489,
"text": "(Nguyen et al., 2016)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Implementation Details",
"sec_num": "3.3"
},
{
"text": "Now we are ready to describe the proposed Talk to Paper application, powered by our SOCO QA framework.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Talk to Paper",
"sec_num": "4"
},
{
"text": "Talk to Paper's data source contains NLP papers published last 3 years in ACL, NAACL, EMNLP and SiGdial in ACL Anthology 5 , which attributes to 3897 papers published in the proceedings of these conferences (we will continuously expand the database by adding more papers from previous years as well as new published papers). We first use SOCO's document parser to extract text data from the PDFs and converted them into the frame format defined in the previous section. Then we use soco-core-python to index the frames and query for answers via its RESTful API endpoint. The indexing process takes about 2 hours.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data Source",
"sec_num": "4.1"
},
{
"text": "Talk to Paper is an web app that can be used on any modern browser. There are three major pages:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "User Interfaces",
"sec_num": "4.2"
},
{
"text": "\u2022 Main search page",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "User Interfaces",
"sec_num": "4.2"
},
{
"text": "\u2022 In-paper search page",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "User Interfaces",
"sec_num": "4.2"
},
{
"text": "\u2022 Annotation page.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "User Interfaces",
"sec_num": "4.2"
},
{
"text": "Main Search page: The main search page is similar to the standard Google-like search interface as shown in Figure 3 , including input search box and query auto completion (based on generated questions from GPT-2).The responding answers will be highlighted in each returned results.",
"cite_spans": [],
"ref_spans": [
{
"start": 107,
"end": 115,
"text": "Figure 3",
"ref_id": "FIGREF2"
}
],
"eq_spans": [],
"section": "User Interfaces",
"sec_num": "4.2"
},
{
"text": "In-paper Search Page: Previously, people search information in the paper by clicking Con-trol+F, which is a well-known shortcut key often used to find text in the current page using the exact character matching or regular expression. It is often used to input a keyword and highlight the matched string and allow to navigate the next matching or previous matching. We provide a similar interface to find the answer inside a specific paper as shown in Figure 2 . Instead of searching information using a keyword, the proposed method allow to find the information using natural language queries. The retrieved answers are highlighted and it is also allowed to navigate next answer or previous answer. It will be useful to find multiple answers in the paper.",
"cite_spans": [],
"ref_spans": [
{
"start": 451,
"end": 459,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "User Interfaces",
"sec_num": "4.2"
},
{
"text": "Annotation Page: We allow to annotate the question and answer spans in the in-paper search page as shown in Figure 4 . All annotated data are visible in the preview page. If a user wants to annotate the data, the user can simply drag the text and write a question. The data will be automatically saved in the database. Unlike other open-domain QA datasets, we cannot ask to crowd workers, students, or part-time contractors to annotate on academic papers because it is hard to annotate without the domain knowledge. Therefore, we will welcome contributions from the research community to make useful resources together for the further research. ",
"cite_spans": [],
"ref_spans": [
{
"start": 108,
"end": 116,
"text": "Figure 4",
"ref_id": "FIGREF3"
}
],
"eq_spans": [],
"section": "User Interfaces",
"sec_num": "4.2"
},
{
"text": "The typical use cases are as following: ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Use Cases",
"sec_num": "4.3"
},
{
"text": "A: Currently, one of the best performing NLI models (e.g., on the SNLI dataset) for three way classification is (Liu et al., 2019) .",
"cite_spans": [
{
"start": 112,
"end": 130,
"text": "(Liu et al., 2019)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Q: What is the best system for NLI?",
"sec_num": null
},
{
"text": "Identification of Tasks, Datasets, Evaluation Metrics, and Numeric Scores for Scientific Leaderboards Construction ... Our work differs in the information extracted and consequently in what context and hypothesis information we model. Currently, one of the best performing NLI models (e.g., on the SNLI dataset) for three way classification is (Liu et al., 2019) . The authors apply deep neural networks and make use of BERT (Devlin et al., 2019) ,... with its previous and next context. The related FAQs are also presented with the \"You may also want to know\" message. The user can also uses filters to narrow down to the answer in one or more specific paper.",
"cite_spans": [
{
"start": 344,
"end": 362,
"text": "(Liu et al., 2019)",
"ref_id": null
},
{
"start": 425,
"end": 446,
"text": "(Devlin et al., 2019)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Title:",
"sec_num": null
},
{
"text": "2. the user clicks the \"view in document\" to check the answer with the original paper. The in-paper search page will be shown. The user can either read the paper or uses in-paper search, e.g. what is the main contribution? to let Talk to Paper auto scroll and highlight relevant answer spans (Figure 2) .",
"cite_spans": [],
"ref_spans": [
{
"start": 292,
"end": 302,
"text": "(Figure 2)",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Title:",
"sec_num": null
},
{
"text": "3. the user may think certain span in the paper contains important information and uses the annotation function to add related questions to this span. This new annotations will be saved in to databases and will be added to the public dataset after manual inspection.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Title:",
"sec_num": null
},
{
"text": "4. the user may uses the dataset as way to train and test performance of a question answering system. The Talk to Paper dataset is different from existing corpus because it contains highly technical text data that are substantially different from Wikipedia, which is a major source of most of the existing QA datasets.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Title:",
"sec_num": null
},
{
"text": "In this section, we first present quantitative preliminary evaluation results the effectiveness of the proposed SOCO-QA framework on a number of standard QA datasets. Then we show results on the data collected from our initial user study.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments and Results",
"sec_num": "5"
},
{
"text": "This preliminary studies focuses on comparison between SOCO-QA against classic BM25 (Robertson et al., 2009) . BM25-based methods remain to be the mainstream methods for document retrieval in industry. Previous work in open domain question answering has shown that BM25 is a difficult baseline to surpass when questions were written by workers who have prior knowledge of the answer (Lee et al., 2019a) . We will leave more comprehensive comparisons against other learning-based methods to future work, since the main goal of this demo paper is to present the system along with its dataset. We use the built-in elastic search (Gormley and Tong, 2015) (NQ) , MS MARCO (Nguyen et al., 2016) and Trivia QA (Joshi et al., 2017) . We break documents from the development set into sentence-level answer frames, and uses the queries in the development set to compute Mean Reciprocal Rank (MRR) and Recall at 5 (R@5) as the evaluation metrics. The data statics are summarized in Table 2 .",
"cite_spans": [
{
"start": 84,
"end": 108,
"text": "(Robertson et al., 2009)",
"ref_id": "BIBREF15"
},
{
"start": 383,
"end": 402,
"text": "(Lee et al., 2019a)",
"ref_id": "BIBREF10"
},
{
"start": 626,
"end": 650,
"text": "(Gormley and Tong, 2015)",
"ref_id": "BIBREF6"
},
{
"start": 667,
"end": 688,
"text": "(Nguyen et al., 2016)",
"ref_id": "BIBREF12"
},
{
"start": 703,
"end": 723,
"text": "(Joshi et al., 2017)",
"ref_id": "BIBREF7"
}
],
"ref_spans": [
{
"start": 971,
"end": 978,
"text": "Table 2",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Results for SOCO-QA performance",
"sec_num": "5.1"
},
{
"text": "Quantitative Results: Table 3 shows the main results. The proposed SOCO-QA model is able to significantly outperform the baseline BM25 on all datasets. The proposed method is particularly powerful on real query data, e.g. NQ and MARCO where the question writer does not the exact answer they are looking for, so that there is often a low word overlapping between the question and the answer. Table 3 shows a striking 251% and 253.6% relative MRR improvement on the NQ and MARCO dataset. On the other hand, SOCO is also able to beat BM25 on SQuAD and Trivia dataset, where there is significant more question-to-answer word lapping.",
"cite_spans": [],
"ref_spans": [
{
"start": 22,
"end": 29,
"text": "Table 3",
"ref_id": "TABREF3"
},
{
"start": 392,
"end": 399,
"text": "Table 3",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Results for SOCO-QA performance",
"sec_num": "5.1"
},
{
"text": "Qualitative Results: to provide better understanding between BM25-based search versus SOCO-QA, the following are some example sideby-side comparisons:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results for SOCO-QA performance",
"sec_num": "5.1"
},
{
"text": "\u2022 SOCO: We compare our baselines with a fine-tuned BERT model (Devlin et al., 2018) .",
"cite_spans": [
{
"start": 62,
"end": 83,
"text": "(Devlin et al., 2018)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Results for SOCO-QA performance",
"sec_num": "5.1"
},
{
"text": "BERT is a pre-trained deep bidirectional transformer model that can encode sentences into dense vector representations. It is trained on large un-annotated corpora such as Wikipedia and the BooksCorpus (Zhu et al., 2015).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results for SOCO-QA performance",
"sec_num": "5.1"
},
{
"text": "\u2022 ES Default (BM25): for the claim pairs with distance values 2 to 5 as shown in Table 3 . We find that BERT model is consistently the best performing model for all distance pairs. As we increase the distance, the models achieve higher prediction performance.",
"cite_spans": [],
"ref_spans": [
{
"start": 81,
"end": 88,
"text": "Table 3",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Results for SOCO-QA performance",
"sec_num": "5.1"
},
{
"text": "The main observations is that BM25 falls short in understanding the intent of the query. Although it is also able to find sentences that are relevant to the query terms, it does not rank sentences that can \"answer\" the query higher. On the other hand, SOCO-QA is able to recognize target answer a query is looking for, e.g. a definition, and rank sentences that can directly resolves the questions higher.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results for SOCO-QA performance",
"sec_num": "5.1"
},
{
"text": "We asked NLP researchers via social network, e.g. Twitter, to try out Talk to Paper and we are able to collect 3137 queries in roughly two weeks. The logged query data and its annotation will soon be made publicly available). Table 1 shows example queries where the system is able to find relevant answers to real user queries. Analysis shows that the most frequent query type were asking about the objectives or the meaning of terms (e.g., what are pretraining objectives, what is LSTM?). Another popular question type is to ask about the state-ofthe-art method to solve a particular problem, e.g. What is the best system for NLI?. We also found that the generated questions that are presented as auto-completion and FAQs are particularly popular. About 51.7% of queries were from the suggested questions. This results is inline with research work in human-computer interaction that utilizes machine intelligent systems to assist human users to better discover knowledge (Lee et al., 2019b) .",
"cite_spans": [
{
"start": 972,
"end": 991,
"text": "(Lee et al., 2019b)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [
{
"start": 226,
"end": 233,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "Data Analysis",
"sec_num": "5.2"
},
{
"text": "We present Talk to Paper, a QA system for NLP papers powered by SOCO-QA. Experiments confirm the effectiveness of our proposed approach and show superior search experience compared to traditional search engine. We welcome contributions from the research community to curate useful resources together for the further research. Future work include (1) expanding the database to more papers (2) improving the QA model using the collected data to better handle question answering in the context of research domain.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "https://docs.soco.ai/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://www.aclweb.org/anthology/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "We would like to acknowledge the joint effort from SOCO's development team, including Haolin Wang, Yanran Han and Omer Riaz to make this work possible.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgments",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Reqa: An evaluation for endto-end answer retrieval models",
"authors": [
{
"first": "Amin",
"middle": [],
"last": "Ahmad",
"suffix": ""
},
{
"first": "Noah",
"middle": [],
"last": "Constant",
"suffix": ""
},
{
"first": "Yinfei",
"middle": [],
"last": "Yang",
"suffix": ""
},
{
"first": "Daniel",
"middle": [],
"last": "Cer",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1907.04780"
]
},
"num": null,
"urls": [],
"raw_text": "Amin Ahmad, Noah Constant, Yinfei Yang, and Daniel Cer. 2019. Reqa: An evaluation for end- to-end answer retrieval models. arXiv preprint arXiv:1907.04780.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Universal sentence encoder",
"authors": [
{
"first": "Daniel",
"middle": [],
"last": "Cer",
"suffix": ""
},
{
"first": "Yinfei",
"middle": [],
"last": "Yang",
"suffix": ""
},
{
"first": "Sheng-Yi",
"middle": [],
"last": "Kong",
"suffix": ""
},
{
"first": "Nan",
"middle": [],
"last": "Hua",
"suffix": ""
},
{
"first": "Nicole",
"middle": [],
"last": "Limtiaco",
"suffix": ""
},
{
"first": "Rhomni",
"middle": [],
"last": "St John",
"suffix": ""
},
{
"first": "Noah",
"middle": [],
"last": "Constant",
"suffix": ""
},
{
"first": "Mario",
"middle": [],
"last": "Guajardo-Cespedes",
"suffix": ""
},
{
"first": "Steve",
"middle": [],
"last": "Yuan",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Tar",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1803.11175"
]
},
"num": null,
"urls": [],
"raw_text": "Daniel Cer, Yinfei Yang, Sheng-yi Kong, Nan Hua, Nicole Limtiaco, Rhomni St John, Noah Constant, Mario Guajardo-Cespedes, Steve Yuan, Chris Tar, et al. 2018. Universal sentence encoder. arXiv preprint arXiv:1803.11175.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Reading Wikipedia to answer opendomain questions",
"authors": [
{
"first": "Danqi",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Adam",
"middle": [],
"last": "Fisch",
"suffix": ""
},
{
"first": "Jason",
"middle": [],
"last": "Weston",
"suffix": ""
},
{
"first": "Antoine",
"middle": [],
"last": "Bordes",
"suffix": ""
}
],
"year": 2017,
"venue": "Association for Computational Linguistics (ACL)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Danqi Chen, Adam Fisch, Jason Weston, and Antoine Bordes. 2017. Reading Wikipedia to answer open- domain questions. In Association for Computa- tional Linguistics (ACL).",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Bert: Pre-training of deep bidirectional transformers for language understanding",
"authors": [
{
"first": "Jacob",
"middle": [],
"last": "Devlin",
"suffix": ""
},
{
"first": "Ming-Wei",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "Kenton",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Kristina",
"middle": [],
"last": "Toutanova",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1810.04805"
]
},
"num": null,
"urls": [],
"raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understand- ing. arXiv preprint arXiv:1810.04805.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Learning to ask: Neural question generation for reading comprehension",
"authors": [
{
"first": "Xinya",
"middle": [],
"last": "Du",
"suffix": ""
},
{
"first": "Junru",
"middle": [],
"last": "Shao",
"suffix": ""
},
{
"first": "Claire",
"middle": [],
"last": "Cardie",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "1342--1352",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xinya Du, Junru Shao, and Claire Cardie. 2017. Learn- ing to ask: Neural question generation for reading comprehension. In Proceedings of the 55th An- nual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1342- 1352.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Building watson: An overview of the deepqa project",
"authors": [
{
"first": "David",
"middle": [],
"last": "Ferrucci",
"suffix": ""
},
{
"first": "Eric",
"middle": [],
"last": "Brown",
"suffix": ""
},
{
"first": "Jennifer",
"middle": [],
"last": "Chu-Carroll",
"suffix": ""
},
{
"first": "James",
"middle": [],
"last": "Fan",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Gondek",
"suffix": ""
},
{
"first": "Aditya",
"middle": [
"A"
],
"last": "Kalyanpur",
"suffix": ""
},
{
"first": "Adam",
"middle": [],
"last": "Lally",
"suffix": ""
},
{
"first": "William",
"middle": [],
"last": "Murdock",
"suffix": ""
},
{
"first": "Eric",
"middle": [],
"last": "Nyberg",
"suffix": ""
},
{
"first": "John",
"middle": [],
"last": "Prager",
"suffix": ""
}
],
"year": 2010,
"venue": "AI magazine",
"volume": "31",
"issue": "3",
"pages": "59--79",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "David Ferrucci, Eric Brown, Jennifer Chu-Carroll, James Fan, David Gondek, Aditya A Kalyanpur, Adam Lally, J William Murdock, Eric Nyberg, John Prager, et al. 2010. Building watson: An overview of the deepqa project. AI magazine, 31(3):59-79.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Elasticsearch: the definitive guide: a distributed real-time search and analytics engine",
"authors": [
{
"first": "Clinton",
"middle": [],
"last": "Gormley",
"suffix": ""
},
{
"first": "Zachary",
"middle": [],
"last": "Tong",
"suffix": ""
}
],
"year": 2015,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Clinton Gormley and Zachary Tong. 2015. Elastic- search: the definitive guide: a distributed real-time search and analytics engine. \" O'Reilly Media, Inc.\".",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension",
"authors": [
{
"first": "Mandar",
"middle": [],
"last": "Joshi",
"suffix": ""
},
{
"first": "Eunsol",
"middle": [],
"last": "Choi",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Daniel",
"suffix": ""
},
{
"first": "Luke",
"middle": [],
"last": "Weld",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Zettlemoyer",
"suffix": ""
}
],
"year": 2017,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1705.03551"
]
},
"num": null,
"urls": [],
"raw_text": "Mandar Joshi, Eunsol Choi, Daniel S Weld, and Luke Zettlemoyer. 2017. Triviaqa: A large scale distantly supervised challenge dataset for reading comprehen- sion. arXiv preprint arXiv:1705.03551.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Natural questions: a benchmark for question answering research",
"authors": [
{
"first": "Tom",
"middle": [],
"last": "Kwiatkowski",
"suffix": ""
},
{
"first": "Jennimaria",
"middle": [],
"last": "Palomaki",
"suffix": ""
},
{
"first": "Olivia",
"middle": [],
"last": "Redfield",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Collins",
"suffix": ""
},
{
"first": "Ankur",
"middle": [],
"last": "Parikh",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Alberti",
"suffix": ""
},
{
"first": "Danielle",
"middle": [],
"last": "Epstein",
"suffix": ""
},
{
"first": "Illia",
"middle": [],
"last": "Polosukhin",
"suffix": ""
},
{
"first": "Jacob",
"middle": [],
"last": "Devlin",
"suffix": ""
},
{
"first": "Kenton",
"middle": [],
"last": "Lee",
"suffix": ""
}
],
"year": 2019,
"venue": "Transactions of the Association for Computational Linguistics",
"volume": "7",
"issue": "",
"pages": "453--466",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tom Kwiatkowski, Jennimaria Palomaki, Olivia Red- field, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, et al. 2019. Natural questions: a bench- mark for question answering research. Transactions of the Association for Computational Linguistics, 7:453-466.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Ranking paragraphs for improving answer recall in open-domain question answering",
"authors": [
{
"first": "Jinhyuk",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Seongjun",
"middle": [],
"last": "Yun",
"suffix": ""
},
{
"first": "Hyunjae",
"middle": [],
"last": "Kim",
"suffix": ""
},
{
"first": "Miyoung",
"middle": [],
"last": "Ko",
"suffix": ""
},
{
"first": "Jaewoo",
"middle": [],
"last": "Kang",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1810.00494"
]
},
"num": null,
"urls": [],
"raw_text": "Jinhyuk Lee, Seongjun Yun, Hyunjae Kim, Miyoung Ko, and Jaewoo Kang. 2018. Ranking paragraphs for improving answer recall in open-domain ques- tion answering. arXiv preprint arXiv:1810.00494.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Latent retrieval for weakly supervised open domain question answering",
"authors": [
{
"first": "Kenton",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Ming-Wei",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "Kristina",
"middle": [],
"last": "Toutanova",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1906.00300"
]
},
"num": null,
"urls": [],
"raw_text": "Kenton Lee, Ming-Wei Chang, and Kristina Toutanova. 2019a. Latent retrieval for weakly supervised open domain question answering. arXiv preprint arXiv:1906.00300.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Learning autocomplete systems as a communication game",
"authors": [
{
"first": "Mina",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Tatsunori",
"suffix": ""
},
{
"first": "Percy",
"middle": [],
"last": "Hashimoto",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Liang",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1911.06964"
]
},
"num": null,
"urls": [],
"raw_text": "Mina Lee, Tatsunori B Hashimoto, and Percy Liang. 2019b. Learning autocomplete systems as a commu- nication game. arXiv preprint arXiv:1911.06964.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Ms marco: a human-generated machine reading comprehension dataset",
"authors": [
{
"first": "Tri",
"middle": [],
"last": "Nguyen",
"suffix": ""
},
{
"first": "Mir",
"middle": [],
"last": "Rosenberg",
"suffix": ""
},
{
"first": "Xia",
"middle": [],
"last": "Song",
"suffix": ""
},
{
"first": "Jianfeng",
"middle": [],
"last": "Gao",
"suffix": ""
},
{
"first": "Saurabh",
"middle": [],
"last": "Tiwary",
"suffix": ""
},
{
"first": "Rangan",
"middle": [],
"last": "Majumder",
"suffix": ""
},
{
"first": "Li",
"middle": [],
"last": "Deng",
"suffix": ""
}
],
"year": 2016,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tri Nguyen, Mir Rosenberg, Xia Song, Jianfeng Gao, Saurabh Tiwary, Rangan Majumder, and Li Deng. 2016. Ms marco: a human-generated machine read- ing comprehension dataset.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Language models are unsupervised multitask learners",
"authors": [
{
"first": "Alec",
"middle": [],
"last": "Radford",
"suffix": ""
},
{
"first": "Jeff",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Rewon",
"middle": [],
"last": "Child",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Luan",
"suffix": ""
},
{
"first": "Dario",
"middle": [],
"last": "Amodei",
"suffix": ""
},
{
"first": "Ilya",
"middle": [],
"last": "Sutskever",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. 2019. Language models are unsupervised multitask learners.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Squad: 100,000+ questions for machine comprehension of text",
"authors": [
{
"first": "Pranav",
"middle": [],
"last": "Rajpurkar",
"suffix": ""
},
{
"first": "Jian",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Konstantin",
"middle": [],
"last": "Lopyrev",
"suffix": ""
},
{
"first": "Percy",
"middle": [],
"last": "Liang",
"suffix": ""
}
],
"year": 2016,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1606.05250"
]
},
"num": null,
"urls": [],
"raw_text": "Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. 2016. Squad: 100,000+ questions for machine comprehension of text. arXiv preprint arXiv:1606.05250.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "The probabilistic relevance framework: Bm25 and beyond",
"authors": [
{
"first": "Stephen",
"middle": [],
"last": "Robertson",
"suffix": ""
},
{
"first": "Hugo",
"middle": [],
"last": "Zaragoza",
"suffix": ""
}
],
"year": 2009,
"venue": "Foundations and Trends R in Information Retrieval",
"volume": "3",
"issue": "4",
"pages": "333--389",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Stephen Robertson, Hugo Zaragoza, et al. 2009. The probabilistic relevance framework: Bm25 and be- yond. Foundations and Trends R in Information Re- trieval, 3(4):333-389.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Real-time open-domain question answering with dense-sparse phrase index",
"authors": [
{
"first": "Minjoon",
"middle": [],
"last": "Seo",
"suffix": ""
},
{
"first": "Jinhyuk",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Tom",
"middle": [],
"last": "Kwiatkowski",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Ankur",
"suffix": ""
},
{
"first": "Ali",
"middle": [],
"last": "Parikh",
"suffix": ""
},
{
"first": "Hannaneh",
"middle": [],
"last": "Farhadi",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Hajishirzi",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1906.05807"
]
},
"num": null,
"urls": [],
"raw_text": "Minjoon Seo, Jinhyuk Lee, Tom Kwiatkowski, Ankur P Parikh, Ali Farhadi, and Hannaneh Ha- jishirzi. 2019. Real-time open-domain question answering with dense-sparse phrase index. arXiv preprint arXiv:1906.05807.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Localitysensitive hashing for finding nearest neighbors",
"authors": [
{
"first": "Malcolm",
"middle": [],
"last": "Slaney",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Casey",
"suffix": ""
}
],
"year": 2008,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Malcolm Slaney and Michael Casey. 2008. Locality- sensitive hashing for finding nearest neighbors [lec- ture notes].",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "The trec-8 question answering track evaluation",
"authors": [
{
"first": "M",
"middle": [],
"last": "Ellen",
"suffix": ""
},
{
"first": "Dawn",
"middle": [
"M"
],
"last": "Voorhees",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Tice",
"suffix": ""
}
],
"year": 1999,
"venue": "TREC",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ellen M Voorhees and Dawn M Tice. 1999. The trec-8 question answering track evaluation. In TREC, vol- ume 1999, page 82. Citeseer.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "End-to-end open-domain question answering with bertserini",
"authors": [
{
"first": "Wei",
"middle": [],
"last": "Yang",
"suffix": ""
},
{
"first": "Yuqing",
"middle": [],
"last": "Xie",
"suffix": ""
},
{
"first": "Aileen",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "Xingyu",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Luchen",
"middle": [],
"last": "Tan",
"suffix": ""
},
{
"first": "Kun",
"middle": [],
"last": "Xiong",
"suffix": ""
},
{
"first": "Ming",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Jimmy",
"middle": [],
"last": "Lin",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1902.01718"
]
},
"num": null,
"urls": [],
"raw_text": "Wei Yang, Yuqing Xie, Aileen Lin, Xingyu Li, Luchen Tan, Kun Xiong, Ming Li, and Jimmy Lin. 2019. End-to-end open-domain question answering with bertserini. arXiv preprint arXiv:1902.01718.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Towards end-to-end learning for dialog state tracking and management using deep reinforcement learning",
"authors": [
{
"first": "Tiancheng",
"middle": [],
"last": "Zhao",
"suffix": ""
},
{
"first": "Maxine",
"middle": [],
"last": "Eskenazi",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue",
"volume": "",
"issue": "",
"pages": "1--10",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tiancheng Zhao and Maxine Eskenazi. 2016. To- wards end-to-end learning for dialog state tracking and management using deep reinforcement learning. In Proceedings of the 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 1-10.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"uris": null,
"num": null,
"text": "Overall workflow of the proposed SOCO framework. The machine reading step in the dashed box is optional.",
"type_str": "figure"
},
"FIGREF1": {
"uris": null,
"num": null,
"text": "In-paper Search Page of Talk to Paper.",
"type_str": "figure"
},
"FIGREF2": {
"uris": null,
"num": null,
"text": "Main search page of Talk to Paper.",
"type_str": "figure"
},
"FIGREF3": {
"uris": null,
"num": null,
"text": "Annotation page of Talk to Paper.",
"type_str": "figure"
},
"FIGREF4": {
"uris": null,
"num": null,
"text": "1. A user asks a question or click one of FAQsin the main search page. The N-best results will be presented with the highlighted answer Examples Paragraphs Q: what are pretraining objectives? A: that pretraining will improve downstream tasks with fine-tuning on the entire available data Title: Pretraining Methods for Dialog Context Representation Learning ... The pretraining objectives are assessed under four different hypotheses: (1) that pretraining will improve downstream tasks with finetuning on the entire available data, (2) that pretraining will result in better convergence, ...Q: what is LSTM?A: Long Short-Term Memory Network Title: Reasoning with Sarcasm by Reading In-between ... The filter width is 3 and number of filters f = 100. LSTM is a vanilla Long Short-Term Memory Network. The size of the LSTM cell is set to d = 100. ATT-LSTM (Attention-based LSTM) is a LSTM model with a neural attention mechanism applied to all the...",
"type_str": "figure"
},
"TABREF0": {
"html": null,
"text": "Example results from real user queries",
"num": null,
"content": "<table/>",
"type_str": "table"
},
"TABREF2": {
"html": null,
"text": "Statistics on the evaluation datasets.",
"num": null,
"content": "<table><tr><td/><td>BM25</td><td>SOCO</td><td/></tr><tr><td/><td colspan=\"2\">MRR R@5 MRR</td><td>R@5</td></tr><tr><td>SQuAD</td><td>58.0</td><td>69.0 60.9</td><td>73.2</td></tr><tr><td>Trivia</td><td>29.0</td><td>38.7 34.0</td><td>59.2</td></tr><tr><td>NQ</td><td>19.7</td><td>25.1 69.3</td><td>87.3</td></tr><tr><td colspan=\"2\">MARCO 20.7</td><td>27.0 73.2</td><td>92.8</td></tr></table>",
"type_str": "table"
},
"TABREF3": {
"html": null,
"text": "Main evaluation results.",
"num": null,
"content": "<table/>",
"type_str": "table"
}
}
}
} |